GQA-MSCG
收藏arXiv2025-09-30 收录
下载链接:
https://github.com/NeverMoreLCH/MSCG
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是在GQA数据集的基础上构建的,旨在定量评估视觉问答模型在多源组合泛化能力方面的表现。它包含了由语言和视觉原语组成的三种新型组合样本。该数据集的任务是视觉问答(VQA)。
Built upon the GQA dataset, this dataset is designed to quantitatively assess the multi-source compositional generalization capabilities of visual question answering (VQA) models. It includes three novel compositional samples composed of linguistic and visual primitives. The core task of this dataset is visual question answering (VQA).
提供机构:
Authors of the paper



